On Facebook, you can create lookalike (LLA) audiences: where Facebook takes your custom audience and finds a pool of similar users. This is extremely useful and using machine learning technology to find those who are interested in your ads and likely to convert.
Facebook lets you choose from 1%-10% in the level of similarity, while Madgicx allows you to go up to 20%.
When launching a LLA audience, Facebook prompts you to choose a lookalike percentage. This percentage is the top portion of the general population who are the most similar to your original target audience.
For example, a 1% Lookalike in the US would mean 1% of the US users population that are most similar to your audience: this audience typically has a reach of about 2 million people (or 20 million for a 10% Lookalike).
The lowest the look-alike percentage is, the most similar it will be to your seed audience (Custom audience), so they are likely to perform better: also, the higher quality of these audiences tends to translate into faster performance feedback. However, are the smallest, so they have the lowest reach: which also means they might saturate faster.
Start off by testing several audiences with a 1% lookalike percentage: if you advertise in smaller cities or countries, maybe start off with 1-3%. Once you’ve found your best lookalike audiences you can start scaling to a much larger budget simply by, relaunching them with larger lookalike percentages: 5%, 10%, 15%, and even 20% if you are aiming at a really big budget.
By the way, the seed audience is being excluded automatically by facebook so there is no need to worry about mixing up the funnel stages
You can use the Madgicx settings to automatically exclude already running lookalike percentages. (Ie. you can launch a 1-5% lookalike excluding the 0-1%).